基于工程经验知识约束神经网络的套管选材推荐模型  

Recommendation Model for Casing Material Selection Based on Empirical Engineering Knowledge Neural Network

作　　者：李奕政 陈掌星 王正 孟洋 张永安 丁瑞辰 LI Yizheng;CHEN Zhangxing;WANG Zheng;MENG Yang;ZHANG Yongan;DING Ruichen(Eastern Institute of Technology,Ningbo,Zhejiang 315200,China;The Hong Kong Polytechnic University,Hong Kong 999077,China;Ningbo Institute of Digital Twin,Eastern Institute of Technology,Ningbo,Zhejiang 315200,China;College of Petroleum Engineering,China University of Petroleum(Beijing),Beijing 102249,China)

机构地区：[1]宁波东方理工大学 [2]香港理工大学,中国香港 [3]宁波数字孪生(东方理工)研究院 [4]中国石油大学(北京)石油工程学院 [5]美国工程院

出　　处：《钻采工艺》2025年第1期29-36,共8页Drilling & Production Technology

摘　　要：科学合理的套管设计对于保障油气生产安全、预防资源浪费与财产损失至关重要,是确保钻井及开采作业顺利进行的核心要素。文章针对套管设计所面临的日渐复杂的工况及数据管理难题,提出了一种工程经验知识约束神经网络(EKNN)的方法,旨在指导套管设计工作。该方法利用现有套管数据资产,基于机器学习的强大数据处理能力,构建一个高效的套管选材推荐模型,通过嵌入套管强度校核知识,提高模型选材推荐的科学性和安全性。首先基于套管数据资产构建套管选材数据集,选取经典的MLP(Multi-layer Perceptron,多层感知机)神经网络建立套管选材推荐模型;采用多任务学习策略设计网络结构,模型在训练过程中优化套管的壁厚及钢级分类;然后以模型预测精度通过超参数优化工具Optuna优化模型超参数;最后以损失函数修正的方式嵌入套管强度校核知识(工程经验知识约束)完成EKNN模型的建立。工程经验知识约束神经网络模型对套管选材的预测精度可达90%以上,模型预测得出的套管选材可以很好地满足套管柱强度设计要求,为各油气田有效利用累积的数据资产、降低成本及优化决策提供有力支持。Scientific and rational casing design is of paramount importance for ensuring the safety of oil and gas production,preventing resource wastage,and avoiding property damage,serving as a core component in facilitating the smooth execution of drilling and extraction operations.In response to the increasingly complex operational conditions and data management challenges encountered in casing design for oil and gas exploration and development,this paper introduces an Engineering-Empirical Knowledge Neural Network(EKNN)approach.This method leverages existing casing data assets and the powerful data processing ability of machine learning to construct an efficient casing material selection recommendation model.By incorporating empirical knowledge of casing strength verification,the model's scientific rigor and safety are enhanced.The process begins with the creation of a casing material selection and design dataset using available casing data.A classic Multi-Layer Perceptron(MLP)neural network is then employed to establish the recommendation model.A multitask learning strategy is adopted in designing the network architecture,enabling the model to simultaneously focus on and optimize both the wall thickness and steel grade classifications of the casing during the training phase.Hyperparameter tuning is conducted using the Optuna optimization tool to maximize the model's predictive accuracy.Finally,the EKNN model is completed by integrating the empirical knowledge of casing strength verification through a modified loss function.The EKNN model achieves a prediction accuracy of over 90%for casing material selection and classification design,with the predicted selections satisfying the strength design criteria for the casing string.This model provides substantial support to oilfield companies in efficiently utilizing their accumulated data assets,reducing costs,and optimizing decision-making processes.

关 键 词：套管设计 工程经验知识约束 神经网络 损失函数修正 

分 类 号：G63[文化科学—教育学]